Introduction
You recognise a friend's face in a fraction of a second, whether they're near or far, on the left or the right of a photo. For a computer, an image is just a big grid of numbers, and teaching it to "see" turned out to be very hard. Convolutional neural networks (CNNs) were the breakthrough that made computer vision work, from phone cameras that find faces to medical scans and self-driving cars.
CNNs are built on a simple observation about images: nearby pixels belong together, and the same pattern (an edge, an eye, a wheel) can appear anywhere in the picture. A CNN looks at small patches of the image at a time and uses the same detector everywhere.
Think of reading a page with a magnifying glass. You slide it across the page, looking at a small area at a time, and you use the same eyes everywhere: you don't need a different way of reading for the top-left corner than for the bottom-right.